Learning and Representing Human Utility
Keywords: Utility, Goal and Action, Preference-based reinforcement learning
Abstract: This essay investigates the critical role of utility both in the process of human decision-making and artificial intelligence system design. We start by examining the studies in behavioural psychology, which demonstrate an early understanding of utility, and list the discussion and applications from philosophy, economy, and sociology. The essay then explores Preference-based Reinforcement Learning (PbRL) as a key computational modeling method of utility. Details of PbRL are carefully discussed, as well as strengths and challenges in existing and future research.
Submission Number: 179
Loading